Horton

Re: pg_dump: fail to restore partition table with serial type

Alvaro Herrera <alvherre@2ndquadrant.com>

View thread View on postgr.es
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Rushabh Lathia <rushabh.lathia@gmail.com>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2019-06-13T03:26:43Z
Lists: pgsql-hackers 
On 2019-Jun-12, Tom Lane wrote:

> Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> > There was indeed one more problem, that only the pg10 pg_upgrade test
> > detected.  Namely, binary-upgrade dump didn't restore for locally
> > defined constraints: they were dumped twice, first in the table
> > definition and later by the ALTER TABLE ADD CONSTRAINT bit for binary
> > upgrade that I had failed to notice.  Ooops.  The reason pg10 detected
> > it and the other branches didn't, is that the only constraint of this
> > ilk that remained after running regress was removed by 05bd889904e0 :-(
> 
> Seems like we'd better put back some coverage for that case, no?

I'll work on that.

> But I'm confused by your reference to 05bd889904e0.  It looks like
> that didn't change anything about tables that weren't getting dropped
> anyhow.

Ah ... yeah, I pasted the wrong commit ID.  That commit indeed removed
one occurrence of constraint check_b, but it wasn't the one that
detected the failure -- the one that did (also named check_b) was
removed by commit 6f6b99d1335b (pg11 only).

Commit aa56671836e6 (in pg10, two months after 05bd889904e0) changed
those tables afterwards so that they wouldn't be dropped. 

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Commits

  1. Make pg_dump emit ATTACH PARTITION instead of PARTITION OF (reprise)

  2. Revert "Make pg_dump emit ATTACH PARTITION instead of PARTITION OF"

  3. Make pg_dump emit ATTACH PARTITION instead of PARTITION OF

  4. Allow a partitioned table to have a default partition.

  5. Give partitioned table "p" in regression tests a less generic name.